The Agent Skills Directory

[DYNAMIC_EXECUTION]: The test suite (tests/test_diversity.py) utilizes importlib.util to dynamically load the diversity_audit.py script. This is a standard and safe development practice for unit testing local components within a package. Additionally, tests/test_review_regressions.py uses __import__('json') to dynamically load the standard Python JSON library for data processing within the tests.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates a diversity audit on user-provided idea data (JSON or plain text), creating a potential attack surface for indirect prompt injection.
Ingestion points: Untrusted content enters the agent's context through the ideas.json file or stdin, which is subsequently processed by scripts/diversity_audit.py.
Boundary markers: The skill documentation (SKILL.md) provides clear instructions on the expected input format, specifically a JSON array or line-separated text.
Capability inventory: The associated Python script is limited to calculating lexical similarity (Jaccard index). It does not possess network capabilities, write permissions to sensitive files, or the ability to execute shell commands based on the input content.
Sanitization: The script performs proactive input validation by enforcing strict resource limits, including a 1MB total input cap, a maximum of 200 ideas, and a 2000-character limit per field. It also verifies data types during the parsing phase to ensure consistency.

generating-novel-ideas